Objective Verification of Clutter Removal and Improved Rainfall Estimates

Efficient removal of clutter and non-meteorological returns is a pre-requisite for accurate rainfall estimation using radar. Many algorithms that recognise ground clutter have been proposed, however evaluating these methods is difficult; often the approach is subjective, involving visual comparison of the radar fields ‘before' and ‘after' clutter removal. A more objective technique is proposed, one which is based on the distribution of observed values of reflectivity, Z, and differential reflectivity, Z_DR, over areas of approximately 5 square km. When clutter is present there is a large scatter of Z and Z_DR values, however after efficient clutter removal plots of Z against Z_DR for the remaining pixels should lie close to a series of well-defined curves.

Raindrop spectra can be represented by an exponential function of the form:

N(D)=N_W exp⁡(D/(D_0 ))

This equation predicts that the plots of Z against Z_DR for a constant N_W should lie on a well-defined curve; changes in N_W lead to proportionate changes in Z and thus a corresponding shift of the curves in the Z direction. Observations show that values of N_W in rain are fairly constant over 5 square km even though individual values of Z and rain rate, R, at each pixel are very variable. An efficient clutter removal algorithm using, for example the texture of differential phase shift, should remove the widely scattered Z and Z_DR pixels deemed to be clutter and leave the pixels classified as rain lying close to a constant N_W curve.

The value of ‘a' in an empirical Z-R relationship of the form ‘Z=aR^b' is inversely proportional to the square root of N_W. The objective analysis of the clutter removal algorithm proceeds as follows: following the clutter removal, the curve of best fit through the Z and Z_DR values gives an estimate of N_W, whilst the goodness of the fit provides an error in N_W. The best clutter removal algorithm is the one that gives the value of N_W with the lowest error. Too aggressive an algorithm will leave too few rainy pixels and a higher error in N_W; too forgiving an algorithm will accept clutter with noisy Z and Z_DR values and a higher error in N_W. Once the optimum N_W has been derived it can be used to define a more appropriate ‘a' in the aforementioned Z-R relationship, allowing a better estimate of rainfall from each individual Z pixel.